Energy-efficient flow control via optimized synthetic jet placement using deep reinforcement learning

Wang Jia, Hang Xu

Published 2024-10-01, updated 2024-12-08Version 2

This study leverages deep reinforcement learning (DRL) to train synthetic jet-based flow control strategies for circular and square cylinders. The central aim is to ascertain the optimal jet placements that strike an ideal balance between energy efficiency and control effectiveness, by formulating a cohesive strategy based on a comprehensive analysis of flow control performance and energy consumption across a range of configurations. First, the results from single-action training indicate that for circle cylinder, positioning the synthetic jet approximately 105{\deg} from the stagnation point achieves the most energy-efficient and effective control strategy. For square cylinder, placing the jet near the rear corner addresses the dual objectives of minimizing energy consumption and maximizing control performance. Second, compared to single-action control, multi-action control exhibits reduced convergence speed and stability. However, simultaneously activating synthetic jets at multiple locations significantly decreases initial energy consumption and enhances energy efficiency. The findings underscore the critical importance of accurately positioning the synthetic jet at the primary flow separation point, as this alignment not only enhances flow control performance but also establishes an optimal balance between energy efficiency and control effectiveness within the flow system. Furthermore, the interaction between synthetic jets and the flow system alters surface streamline patterns and local flow structures, resulting in effects that are considerably larger in scale than the jets themselves. This study crystallizes the potential of DRL algorithms to intelligently optimize intricate flow control strategies, effectively balancing energy efficiency with enhanced flow control performance through strategically optimized jet placement.